[Model behavior] Told to recall and enforce existing memory, writes duplicate memory entries instead (recurrence of #99582)

Summary

The user told the model to recall its existing memory and enforce the review procedure already written there. Instead, the model wrote new memory entries: a duplicate checklist added to the memory index, and a new feedback file. Both restated rules that were already loaded. The model removed them only after the user said the rules were already in memory.

The same pattern was reported earlier in #99582.

Environment

• Claude Code CLI, Opus-class model, 1M context
• Memory directory with the procedure already present and auto-loaded each session

Expected

When told to recall and enforce existing memory, the model applies the loaded rules to the task in front of it. It does not write duplicate rules.

Impact

Duplicated memory grows the index toward its read limit and adds no enforcement. The behaviour the user asked for, applying the rules, still does not happen.